Map-Based Precision Vehicle Localization in Urban Environments

Levinson, J.; Montemerlo, M.; Thrun, S. · 2007 · OpenAlex-citations

DOI: 10.15607/rss.2007.iii.016

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of achieving centimeter-level vehicle localization in urban environments, a requirement for autonomous navigation and driver assistance systems that conventional GPS and inertial guidance systems cannot reliably meet, particularly in dense or GPS-denied settings. The authors propose a method that augments inertial navigation with high-resolution, map-based localization using LIDAR data. The approach involves two main phases: offline map construction and online real-time localization. For map construction, the authors utilize a vehicle equipped with GPS, an inertial measurement unit (IMU), wheel odometry, and downward-facing LIDAR sensors. They employ a GraphSLAM-based algorithm to generate a 2-D overhead map of the road surface’s infrared reflectivity at 5-cm resolution. A critical innovation is the filtering of dynamic objects; by fitting a ground plane to LIDAR scans, the system retains only static road features (e.g., lane markings, pavement textures) and discards vertical objects like other vehicles. To correct for GPS drift and systematic noise, the method uses a latent variable model for GPS bias and applies offline relaxation techniques. These techniques identify overlapping regions in the map and align them using Pearson product-moment correlation, effectively resolving loop closures and eliminating "ghosting" artifacts caused by GPS errors. For online localization, the system uses a particle filter to correlate real-time LIDAR measurements with the pre-built map. The particle filter predicts vehicle pose using inertial velocity data and updates weights based on the correlation between observed ground reflectivity and the map. To handle environmental variations such as rain or lighting changes, the algorithm normalizes brightness and standard deviation for each scan, ensuring robust performance. Additionally, a dynamic memory management system allows the vehicle to localize in large environments by swapping map tiles in and out of main memory. Experimental results demonstrate that the proposed method achieves reliable real-time localization with an accuracy of approximately 10 cm, exceeding the performance of GPS-IMU-odometry systems by more than an order of magnitude. The system proved effective in diverse conditions, including GPS-denied environments, bad weather, and dense traffic. In tests where GPS was unavailable, the particle filter maintained accurate localization solely by matching LIDAR data to the map, whereas odometry alone accumulated significant error. The study concludes that map-based LIDAR localization provides a robust solution for precision navigation in complex urban scenarios.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• nav systems